Could ChatGPT replace Human Professionals?
Exploring the potential of AI-powered chatbots to augment or replace human professionals
Context
ChatGPT has been getting a lot of lime-light recently for its promising advancement in the field of Natural-Language Processing (NLP). ChatGPT is a chatbot developed by OpenAI. It stands for Chat Generative Pre-trained Transformer and is built on top of OpenAI’s GPT-3.5 family of large language models.
Introduction from the horse’s mouth:
“I am a computer program called an artificial intelligence or AI. Specifically, I am a large language model trained by OpenAI. My purpose is to assist users by answering questions, providing information, and engaging in conversation. I am able to process and generate human-like text, but I do not have personal thoughts, feelings, or experiences. I exist solely to provide assistance to users. Is there anything else you would like to know?” — ChatGPT
ChatGPT Analysis
I learned about ChatGPT from my colleagues and was very interested to try it out. Here is a brief analysis based on my initial experience with ChatGPT when using the online trial version.
1. Able to determine meaning in context.
One of NLP’s challenges has been understanding references and co-references as well as being able to crack meaning in context. ChatGPT does a really good job at understanding it.
Question: Can you beat humans?
ChatGPT: As an artificial intelligence, I am not capable of competing with humans in the same way that a person might. I do not have personal thoughts, feelings, or experiences, and I am not able to engage in activities in the same way that a human can. I am a computer program designed to process and generate text, and I do not have the ability to engage in physical activities or to interact with the world in the same way that a human can.
However, AI systems like myself can be very good at performing certain tasks that are difficult for humans, such as analyzing large amounts of data or processing large volumes of information quickly. In some cases, AI systems may be able to outperform humans on certain tasks, but this does not mean that they are generally superior to humans. AI systems have their own strengths and limitations, and they are best suited to certain types of tasks and applications.
Notice how in the above question, “beat” can mean both strike and defeat. But ChatGPT was able to quickly grasp the context without needing further clarification. To further verify if my analysis was correct, I asked it explicitly:
Question: What do I mean by “beat” in the previous question?
ChatGPT: In the context of your previous question, it is likely that you were using the word “beat” to mean “perform better than” or “outdo.” When used in this way, the word “beat” suggests competition or comparison between two or more individuals or groups, with one being considered superior to the others.
In the case of your question, it seems that you were asking whether I, as an artificial intelligence, am able to outperform humans in some way. As I mentioned in my previous response, AI systems like myself can be very good at performing certain tasks that are difficult for humans, such as analyzing large amounts of data or processing large volumes of information quickly. However, AI systems have their own strengths and limitations, and they are not generally superior to humans in all areas. Humans have a wide range of capabilities and skills that are not easily replicated by artificial intelligence, and it is important to recognize and value these unique strengths.
2. Able to see past human error in conversations.
It is also able to go one step further and understand both abbreviations as well as typo(s).
Question: how fnny do you think u r?
ChatGPT: As an artificial intelligence, I do not have personal thoughts, feelings, or experiences, and I am not able to evaluate my own sense of humor. I do not have the ability to find things funny or amusing in the same way that a human does.
Instead, my primary function is to process and generate text based on the input I receive. I am able to generate jokes and other forms of humorous content based on the patterns and structures that I have learned from the data I have been trained on, but my ability to create successful jokes depends on the quality of the data and the patterns it contains, as well as the context in which the joke is being told and the audience it is being told to.
It is important to recognize that what one person finds funny may not be amusing to another, and that the success of a joke can depend on a wide range of factors, including the audience’s background, experiences, and sense of humor.
3. Able to process multiple questions in one line.
It is able to comprehend as well as answer two questions at once.
Question: what is the color of the sky? What day is today?
ChatGPT: I’m sorry, but I am unable to determine the color of the sky or the current day as I do not have access to external information or the ability to perceive the world around me. As an artificial intelligence, I am able to process and generate text based on the input I receive, but I do not have the ability to observe or interact with the physical world in the same way that a human does.
I am here to assist you by answering questions, providing information, and engaging in conversation. If you have any other questions or topics you would like to discuss, please let me know and I will do my best to help.
Notice how it responds to both the questions in one sentence, just like a human! That being said, it did fail to give a proper answer. But I understand that it is still in its early stages and is running offline.
4. Able to comprehend questions without explicit words.
It is able to detect that I am asking a question even when I am not using explicit words. It knows that I am trying to compute something and diligently provides me with the correct answer.
Question: 2 + 2 = ?
ChatGPT: 2+2 = 4.
Question: 3+3 ??
ChatGPT: 3+3 = 6.
Question: 3 multiplied by 3
ChatGPT: 3 multiplied by 3 is equal to 9.
5. Unable to continue conversation.
As a chatbot, it fails to keep the conversation rolling. It always expects a question or a command. And when not provided with one, it is confused.
Question: Can you answer in simply yes or no?
ChatGPT: In general, I am able to provide answers to questions in the form of a simple “yes” or “no” if the question can be accurately and completely answered in this way. However, it is important to note that some questions may be complex or may require additional context or explanation in order to be fully understood, and in these cases, a simple “yes” or “no” answer may not be sufficient.
If you have a specific question that you would like me to answer with a simple “yes” or “no,” please let me know and I will do my best to provide an accurate and complete response.
Question: yes
ChatGPT: I’m sorry, but I am unable to provide a meaningful response to your question without more context. Could you please provide more information or clarify your question so that I can better understand what you are asking? I will do my best to assist you to the best of my abilities.
6. Unable to create pleasant looking UI.
Although the concept of best UX is often subjective, yet a simple application created by ChatGPT is not quite pleasant to look at. But that being said, it does do a good job at creating the barebones of an app.
7. Unable to pin-point the oldest dataset date.
Although ChatGPT is informed of its knowledge cut-off date (September 2021), it is unable to give a similar date estimate for its oldest dataset date.
8. Unable to take multiple or choppy commands.
Although it is able to respond to multiple questions at once (see #3), ChatGPT is unable to take choppy commands as input. One must wait until the first command is finished processing until the second command/chat text can be sent. It makes it less fluid an experience in my opinion. For instance, if I were texting someone instructions for a project, and then I realized I either sent them a wrong step or changed my mind, I would quickly send them a correction text without waiting for them to finish the wrong instruction. That is not the case here. If you give wrong input in the first instruction, you must wait until it has responded back to you before you can proceed, at least in the trial version.
9. Gets slower, slowly.
The longer the conversation goes, the slower the ChatGPT gets. I had expected it to get slower only for complex tasks, not for normal conversations. Hence it came out as surprise. Multiple developers in OpenAI discord also complained about similar experiences.
Afterthoughts
In this hyper-reactive world, it is no surprise that anytime a new AI model shows advancement, the society gets a takeover-scare. Whether it be LensAI or ChatGPT, it is important to be excited with the advancements; to be curious and not be scared of it. That doesn’t mean to not be critical of it either. In fact, the CEO of OpenAI has said:
“ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness. It’s a mistake to be relying on it for anything important, but a preview of progress. We have lots of work to do on robustness and truthfulness.” — Sam Altman, CEO of OpenAI
In my personal opinion, I think ChatGPT stays true to its said-purpose, that is, being an assistant. It is not yet capable to fully replace the Human Professionals.
While the fierce debate of ethics, originality, and reliability continues to be a hot-topic, in my opinion, the fascinating possibility that AI holds is hotter. And in this winter, I could really use a summer! What about you?
Thank you for reading this article. If you enjoyed my content, please consider following me to show your support for my work.
